A Research On Network Infrastructures For Data Centers

The research on complex networks has gained a wide range attraction which can be traced up to 1959 where first intruduced as Erdos-Renyi model.Along with the growing of Internet in 90s,and the researching on human society,communication networks and biological networks,the smallworld and scalefree networks has been discovered as an universality of phenomenon and attract many interests.At the meantime,cloud computing expedited the requirement for mass data pro-cessing and the construction of data center.The low cost inherent characteristics in cloud computing evaluate trade-offs for quantity with quality and redundancy with re-liability.How to build the network infrastructure,which provides high performance,high availability and high flexibility,becomes an essential problem of cloud computing data center construction.Based on above understanding,we consider the issue of cloud computing network infrastructure,and conduct the in-depth study on network models over complex net-works,in terms of theory,algorithm and implementation.The researching work and contribution can be summarized as follows:1.Based on the idea of multi-hash,introduced WarpNet network model,which employ nodes as micro routers to connect multiply hash divisions.The theoretical perfor-mance,emulational analysis and experiments over 1000 nodes demonstrated the improvement in network scalibility,flexibility and all-to-all communication band-width.2.Introduced DS smallworld and the construction procedure of SIDN model under the constraint of maximum node degree.Analyzed the average path length,bandwidth and robustness of SIDN network.In terms of theoretical and emulational analysis shows fine extensibility and robustness,which has nearly linear correlation to node scale in extensibility and reversed node drop rate in robustness.3.Analyzed theoretical issues of scalefree network under the condition of maximum degree constraint,and introduced TPSF scalefree network based on recursive con-struction procedures.Analyzed the influence of node/switch proportion variation,and optimizated the proportion parameter.In term of theoretical and emulational analysis shows low cost and high performance-price ratio for cost sensitive sce-nario.4.Analyzed general character:in network model researching,desiged and constructed a general emulation platform named as VMNS,which is based on massive UML virtual machines.Such approach of massive VMs provide identical virtual environ-ment to real world,and the emulation result has linear correlation to real operation situation.VMNS provides a general way for the evolution of network model design works.